Packed DXTn Decoder

/*
Copyright (C) 2013 by Brendan G Bohannon
Email: cr88192@gmail.com
Copying: http://pastebin.com/iJxtZHm6
 */

/*
DXTn packed images.

Each block tag will be encoded in the form (byte):
0 <block:QWORD>     Literal Block.
1-127               Single byte block index.
128-191 X           Two byte block index (16384 blocks).
192-223 XX          Three byte block index (2097152 blocks).
224-238 I           LZ/RLE Run (2-16 blocks, Index)
239     LI          LZ/RLE Run (Length, Index)
240     XXX         24-Bit Index
241     XXXX        32-Bit Index
242-246             Literal Blocks (2-6 Blocks)
247     L           Literal Blocks (L Blocks)
248-255             Reserved

The block index will indicate how many blocks backwards to look for a matching block (1 will repeat the prior block).

Length/Index values will use the same organization as above, only limited to encoding numeric values.

0-127               0-127.
128-191 X           128-16383.
192-223 XX          16384-2097151.
240     XXX         24-Bit Index (0-16777215)
241     XXXX        32-Bit Index (0-4294967295)

Note that DXT5 images will be split into 2 block-planes, with the first encoding the alpha component, followed by the plane encoding the RGB components.
*/

byte *BGBBTJ_PackBCn_DecodeValue(byte *cs, int *rval)
{
    int op, i;

    op=*cs++;
    if(op<128)
        { *rval=op; return(cs); }
    if(op<192)
    {
        *rval=((op-128)<<8)|(*cs++);
        return(cs);
    }
    if(op<224)
    {
        i=((op-192)<<8)|(*cs++);
        i=(i<<8)|(*cs++);
        *rval=i;
        return(cs);
    }

    if(op==240)
    {
        i=*cs++;
        i=(i<<8)|(*cs++);
        i=(i<<8)|(*cs++);
        *rval=i;
        return(cs);
    }
    if(op==241)
    {
        i=*cs++;            i=(i<<8)|(*cs++);
        i=(i<<8)|(*cs++);   i=(i<<8)|(*cs++);
        *rval=i;
        return(cs);
    }

    return(cs);
}

byte *BGBBTJ_PackBCn_DecodeRun(byte *cs, int *ridx, int *rlen)
{
    int op, i;

    op=*cs++;
    if(!op)
        { *ridx=0; *rlen=0; return(cs); }
    if(op<128)
        { *ridx=op; *rlen=1; return(cs); }
    if(op<192)
    {
        *ridx=((op-128)<<8)|(*cs++);
        *rlen=1;
        return(cs);
    }
    if(op<224)
    {
        i=((op-192)<<8)|(*cs++);
        i=(i<<8)|(*cs++);
        *ridx=i;
        *rlen=1;
        return(cs);
    }
    if(op<239)
    {
        cs=BGBBTJ_PackBCn_DecodeValue(cs, ridx);
        *rlen=(op-224)+2;
        return(cs);
    }
    if(op==239)
    {
        cs=BGBBTJ_PackBCn_DecodeValue(cs, rlen);
        cs=BGBBTJ_PackBCn_DecodeValue(cs, ridx);
        return(cs);
    }
    if(op==240)
    {
        i=*cs++;
        i=(i<<8)|(*cs++);
        i=(i<<8)|(*cs++);
        *ridx=i;
        *rlen=1;
        return(cs);
    }
    if(op==241)
    {
        i=*cs++;            i=(i<<8)|(*cs++);
        i=(i<<8)|(*cs++);   i=(i<<8)|(*cs++);
        *ridx=i;
        *rlen=1;
        return(cs);
    }

    if(op<247)
    {
        *ridx=0;
        *rlen=(op-242)+2;
        return(cs);
    }
    if(op==247)
    {
        *ridx=0;
        cs=BGBBTJ_PackBCn_DecodeValue(cs, rlen);
        return(cs);
    }

    *ridx=0; *rlen=0;
    return(cs);
}

byte *BGBBTJ_PackBCn_DecodeBlockArray(byte *ibuf, byte *blks,
    int count, int stride)
{
    byte *cs, *ct, *cte, *cs1, *ct1, *cs1e;
    int bi, bl;
    int i, j;

    cs=ibuf; ct=blks; cte=ct+count*stride;
    while(ct<cte)
    {
        cs=BGBBTJ_PackBCn_DecodeRun(cs, &bi, &bl);
        if(bi==0)
        {
            if(bl>1)
            {
#if defined(X86) || defined(X86_64)
                cs1e=cs+bl*8;
                while(cs<cs1e)
                {
                    *(u64 *)ct=*(u64 *)(cs);
                    cs+=8; ct+=stride;
                }
#else
                for(i=0; i<bl; i++)
                    { memcpy(ct+i*stride, cs+i*8, 8); }
                cs+=bl*8;
                ct+=bl*stride;
#endif
                continue;
            }
#if defined(X86) || defined(X86_64)
            *(u64 *)ct=*(u64 *)cs;
#else
            memcpy(ct, cs, 8);
#endif
            ct+=stride; cs+=8;
            continue;
        }

#if defined(X86) || defined(X86_64)
        cs1=ct-bi*stride;
        cs1e=cs1+bl*stride;
        while(cs1<cs1e)
        {
            *(u64 *)ct=*(u64 *)(cs1);
            cs1+=stride; ct+=stride;
        }
#else
        for(i=0; i<bl; i++)
            { memcpy(ct+i*stride, ct+(i-bi)*stride, 8); }
        ct+=bl*stride;
#endif
    }
    return(cs);
}

BGBBTJ_API int BGBBTJ_PackBCn_DecodeBlocksDXT5(
    byte *ibuf, byte *blks, int count)
{
    byte *cs;
    cs=BGBBTJ_PackBCn_DecodeBlockArray(ibuf, blks, count, 16);
    cs=BGBBTJ_PackBCn_DecodeBlockArray(cs, blks+8, count, 16);
    return(cs-ibuf);
}